text stringlengths 0 2.02k |
|---|
\[01:07:49.07\] So if we could re-envision what the web could be, if we architected the web and created new protocols that were secure first, how game-changing could that be? Those would be great conversations to start having, but first we have to stop arguing about basic stuff. |
So with that said... Actually, I do have a security question for you before we end. Do you know how with password managers you're always copying things onto your clipboard? I always find that a liability, because it's not a one-time copy... So I'm just curious if you guys have ever considered working with browser opera... |
**Mitchell Cohen:** That's actually a feature of 1Password. |
**Amal Hussein:** OMG. |
**Mitchell Cohen:** Even in our new modern web-based frontends we use the system APIs to do that most effectively. So on macOS we actually use the secure clipboard, and clear it after a time-out; we even do this on Linux and Windows, in sort of native ways. |
**Amal Hussein:** That's amazing. |
**Andrew Beyer:** And on iOS and mobile platforms as well. Basically, that is one of the reasons we always go out of our way to support those APIs. And I'll be honest, that's actually a web extension API I would love to see, because we don't have one of those from the browser extension... But in the browser extension, ... |
**Amal Hussein:** That's awesome. Well, I'll tell you, you gained one customer today, I'll tell you that much. It's perfect timing. |
**Andrew Beyer:** One by one. That's how we grew. |
**Amal Hussein:** Yeah. I'm due for my LastPass renewal, so I'm sorry, LastPass; you guys have been great, but... Time for something new. It's been a pleasure having you all today... Seriously. Thank you so much. |
**Andrew Beyer:** Yeah, absolutely. Thank you for having us. I will do one quick call-out, which is if anything we said sounded cool or something you're interested in, we are definitely hiring. I am looking for web developers. If you know TypeScript, you wanna come join us, just check out our Jobs page on 1Password.com... |
**Amal Hussein:** Awesome. And where can people find you all online |
**Andrew Beyer:** You can find me on Twitter at @firebeyer. That's basically the only social media platform I use. I'm not a huge social person. Even LinkedIn, it just makes you a spearfishing target... So I've deleted pretty much every other social media platform... But you can find me there if you wanna chat or set s... |
**Mitchell Cohen:** I am also really only on Twitter, @mitchchn. I've enjoyed all the conversation there about 1Password 8 and participated in it, so please feel free, hit me up with what you like, what you don't like, what you disagree with, what I said on this show... That's great; I really love this conversation, an... |
**Amal Hussein:** Yeah. |
**Andrew Beyer:** Mitch does funny tweets of 1Password spinning on the desktop, because people thought that Electron couldn't do shaking, like we had in the old app... So some of his content is really funny to watch. |
**Mitchell Cohen:** I also do real work, by the way... \[laughs\] |
**Andrew Beyer:** He does some real work, by the way... And I will give a quick shout-out, if you are an iPhone user, iPad user, on Monday iOS 15 comes out, and 1Password will have, I'm hoping, the best web extension there, so you can see what it's like to run 1Password as a web app on an iOS device, which is pretty gr... |
**Amal Hussein:** Yeah, that's so cool. Thank you so much for listening to your customers, and thank you so much for helping drive really good decisions, and obviously, I would say, world-class user experiences. I think a lot of product companies, regardless of what they're doing for their customers, I think could take... |
**Andrew Beyer:** Awesome. Thanks. |
**Mitchell Cohen:** Thank you. |
• Eran Yahav's background as CTO and co-founder of Tabnine and professor of CS at Israel's Technion University |
• His work at IBM in the T.J. Watson Lab on program synthesis for synthesizing low-level concurrent programs |
• The connection between his work on program synthesis and Tabnine's AI-assisted development workflows |
• The challenges of bridging the gap between research and real-world applications |
• Program synthesis as a concept and its relationship to AI and language models |
• The role of deep learning models in accomplishing program synthesis |
• The importance of computational power and availability of training data for program synthesis |
• The current state of program synthesis and its applications in AI-assisted development workflows |
• Art synthesis discussed as a related concept to program synthesis |
• Tabnine AI assistant explained as a tool that generates code based on context |
• Evolution of Tabnine's technology from an early version that used machine learning to generate code from a prompt |
• Challenges of human-machine interaction in program synthesis, including the need for human intent and navigation of possible futures |
• Potential future of program synthesis, including the possibility of higher-level abstraction and user-story-level input |
• Discussion of the nature of programming, including the idea that discovering the spec is a major part of the job |
• Tabnine's role in simplifying the programming process by removing syntactic barriers and making code generation easier |
• Development workflow and AI-assisted coding |
• Feedback loop for user interactions and code samples |
• Tabnine's approach to user privacy and code custody |
• Challenges of gaining user trust and establishing a loyal user base |
• The role of variable reward and user engagement |
• The evolution of AI-assisted development and the emergence of multiple platforms |
• Competition and validation in the AI-assisted development market, specifically the relationship between Tabnine and Microsoft's GitHub Copilot |
• Microsoft's potential impact on Tabnine's market |
• Differentiation of Tabnine from Copilot through personalized model training |
• Challenges of competing with a large company like Microsoft |
• Tabnine's approach to staying competitive and winning customers |
• Future plans and preparations for increased competition |
• Tabnine's financial growth and funding rounds |
• Tabnine's growth and adoption is organic, with no top-down sales or salespeople, and is driven by developer love and word-of-mouth. |
• The company's model is trained on permissive open-source licenses only, excluding GPL and other licenses to avoid potential legal implications. |
• Tabnine's dataset differs from others, such as OpenAI's Codex, which uses a broader range of source code, including GPL. |
• Users have control over their data and inference, allowing them to train their own models on specific datasets and projects. |
• The company is avoiding potential lawsuits by excluding GPL code, but may revisit this policy if the legal landscape changes. |
• Tabnine's focus is on generating value for developers in a way that respects their needs and practices. |
• OpenAI's Codex vs Tabnine |
• Trade-offs with universal models and data usage |
• License and ownership of code used in models |
• Granularity of code completions and human interaction |
• Tabnine's approach to line-by-line completions and guided walkthroughs |
• Differentiating features of Tabnine vs open-source models |
• Eran Yahav discusses the potential for AI assistants to generate code without human input, but notes that it's currently more feasible for idiomatic tasks. |
• The limitations of current AI models in generating code, including the need for human refinement and the importance of context and intent. |
• The role of human psychology in interacting with AI assistants, including the need for an "illusion of control" and the ability to make choices. |
• The importance of user interface design in AI tools, including minimizing cognitive load and providing a seamless experience. |
• The concept of "cognitive load" and its impact on user experience, including the need to avoid overwhelming users with information. |
• The potential for AI assistants to be used as a tool for learning and discovery, rather than simply generating code. |
• The concept of "Bob" as a symbol of AI's potential but limited ability to assist humans |
• The tension between the UX of developer interactions with AI and the AI's own limitations and desires to assist |
• The importance of optimizing AI interactions to avoid overwhelming developers and instead accelerate their work |
• The need to throttle AI to prevent it from "killing" the human with too much information |
• The focus on developing a tool that supports human developers, rather than a model that simply provides knowledge |
• The complexity of integrating AI into various IDEs and editors, with each having its own UX and idiosyncrasies |
• The shared infrastructure and engine between Tabnine's integrations, with some customizations per editor |
• The process of rolling out changes to the AI engine and its interactions with developers across various platforms |
• Tabnine's use of Rust for its ML inference engine and other infrastructure |
• Eran Yahav's enthusiasm for Rust and its ecosystem |
• Limited open-sourcing of Tabnine's code, due to its complexity and lack of cleanliness |
• Company size: around 40 people, mostly engineering, with a focus on Rust and other languages |
• Hiring for engineering positions, with a preference for Rust developers |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.